# Kokon [Health: Active]

**Category:** 💻 Developer Tools  
**Repository:** https://github.com/GabrieleLozupone/kokon  
**GitHub Stars:** 1  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/kokon

## Description
Generate and edit images offline with Qwen-Image 2.1 on your GPU. Needs the Kokon desktop app.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "kokon": {
    "command": "npx",
    "args": ["-y","kokon-mcp"]
  }
}
```

## Documentation

## What Kokon does

The Kokon MCP server gives an AI agent access to local image generation and editing through the Kokon desktop application. It is designed for describing an intended change rather than configuring a large set of image-processing controls. Kokon can work with new images and existing photos, including edits such as removing or replacing an object, changing lighting or style, modifying text, and expanding the canvas.

The application uses Qwen-Image 2.1 and processes work on the local computer. Images and prompts do not need to be uploaded to a hosted service. The application also supports full-size results, keeps the source aspect ratio, and can open iPhone HEIC photos, including HDR content.

## How it works

Install the native Kokon application, then allow its first-run setup to identify the available graphics hardware or Mac memory. Kokon selects a suitable configuration and downloads the required model files. The model download is about 12 GB on first run, and the application verifies the downloaded files.

The MCP server is included with the desktop app. An agent such as Claude or Cursor can send image-generation and image-editing requests to the running local application. Kokon also provides a local HTTP API for scripts, so the same local image workflow can be used without sending the image to a remote service.

## Setup and configuration

Download the Windows installer from the latest release, or use the Apple Silicon macOS disk image marked as beta. The Windows installer is per-user and does not require administrator privileges. On macOS, the application may need to be allowed manually in Privacy & Security because the release is not notarized by Apple.

On first launch, Kokon recommends a preset based on the detected hardware and downloads the models it needs. Supported hardware includes NVIDIA GPUs with at least 12 GB of VRAM. AMD and Intel Arc support are marked beta and also call for at least 12 GB of VRAM. Apple Silicon support is beta, with 32 GB of unified memory recommended. Linux downloads are listed as coming soon.

## Tools and capabilities

The Kokon MCP server supports agent-driven workflows for:

- Generating images from text descriptions.
- Editing selected areas of a photo.
- Removing, replacing, changing, or adding visual elements.
- Relighting or restyling an image.
- Changing visible text in an image.
- Extending an image beyond its original canvas.

Kokon uses a painted area or other editing context to indicate where the model should work, helping preserve unaffected parts of the photo. The application can upscale results with an AI upscaler rather than simple blurry resizing.

## Limitations and notes

Kokon requires its desktop application; the MCP connection is not presented as an independently hosted cloud service. It also requires suitable local hardware and enough storage for the model download. Windows 10 and 11 are supported, while the listed macOS build targets macOS 26 or newer and Apple Silicon in beta. The README does not describe environment variables or a separate command-line installation for the MCP server.

The application performs processing locally and does not require an account. After setup, it checks once daily for a new application version, and that check can be disabled. A private mode is available when the application should retain nothing.

_Full upstream README: https://allmcps.com/mcp/kokon/readme_